Vanishing Spectrometers: How Near-Infrared Instrumentation Got Small, Smart, and Everywhere

Abstract

Near-infrared (NIR) spectroscopy has quietly become one of analytical chemistry's most reengineered techniques. Coverage over the past five years documents a steady move away from bulky, grating- or interferometer-based bench instruments toward wafer-scale photonic chips, resonant-cavity detector arrays, and MEMS interferometers that pack laboratory-grade capability into millimeter-scale packages.^1–5 At the same time, artificial intelligence has migrated from a post-processing afterthought into a core instrument component, cutting calibration burdens, stabilizing bioprocesses in real time, and letting NIR sensors interpret spectra no chemist labeled by hand.^6–8 This article surveys the instrumentation developments most likely to reshape how, where, and by whom NIR measurements are made, from pocket-sized chemical sensors to standoff imaging systems and self-monitoring bioreactor probes, and considers what still separates these devices from routine, trusted use.

Introduction

For most of its history, NIR spectroscopy has been a technique defined by its hardware: a light source, a dispersive or interferometric element, and a detector, all bolted to an optical bench heavy enough to resist vibration. That formula produced excellent chemistry but poor portability. Over the last five years, the balance has flipped. Spectroscopy and the primary literature it draws from now describe NIR instruments built around wafer-scale semiconductor fabrication, microelectromechanical systems (MEMS), and machine-learning models trained to do the work that used to require a spectroscopist's judgment.^1,4,5 The result is not simply smaller versions of familiar instruments; it is a rethinking of what an NIR “spectrometer” even needs to measure to be useful. Some of the most interesting new devices barely resolve a spectrum at all, relying instead on a handful of broad spectral channels and a well-trained model to extract the answer. Below, five threads from recent instrumentation coverage illustrate where the field is headed.

Spectrometers on a Chip: Wafer-Scale Photonics Replace the Grating

The clearest break from tradition comes from fully integrated photonic sensor chips. One recent design pairs a millimeter-scale array of resonant-cavity-enhanced indium gallium arsenide (InGaAs) photodetectors, each pixel tuned to a different broad spectral band across 850–1700 nm, with a handheld module that skips spectral reconstruction entirely.^1,2 Rather than recording a continuous spectrum, the 16-pixel chip feeds its raw photocurrents directly into a partial least squares model, and the approach has already been validated for quantifying fat content in raw milk and sorting plastics by resin type, both with accuracy rivaling conventional benchtop instruments despite a chip footprint of about 2 × 2 mm.^1,2 A parallel approach pushes resolution rather than footprint: a silicon-nitride photonic chip built around cascaded, electrically tunable micro-ring resonators covers a 520-nm window with sub-8-picometer resolution, a bandwidth-to-resolution ratio more than an order of magnitude beyond prior miniaturized designs. It has already demonstrated benchtop-grade glucose detection limits on a chip smaller than a fingernail.³ Together, these devices suggest that wafer-scale fabrication, not incremental shrinking of grating optics, is what finally makes NIR sensing cheap enough for consumer and embedded use.

MEMS, Michelson, and the Miniaturization Arms Race

Even where chip-scale detection has not fully arrived, conventional dispersive and interferometric designs keep shrinking. The 2025 instrumentation roundup from Spectroscopy catalogs an entire generation of field-ready NIR products: an updated MEMS Fourier-transform infrared platform with a smaller footprint and faster acquisition, a vis-NIR field instrument aimed at agriculture and geochemistry, and paired end-user and OEM analyzers offering near-maintenance-free operation alongside enhanced-resolution, actively cooled optics for shorter acquisition times.⁴ That enthusiasm has not been free of hype, however. A widely read comparison of commercial handheld platforms, built on linear variable filters, digital micromirror arrays, Fabry-Perot interferometers, and miniaturized Michelson interferometers, found real trade-offs behind the marketing: instruments with wide spectral coverage generally won at qualitative identification, while high signal-to-noise designs with narrower ranges produced better quantitative calibrations, and none of them were yet ready to be handed, unsupervised, to a non-expert user.⁵ That caution remains a useful corrective to the more enthusiastic instrumentation coverage.

When Software Becomes the Instrument: AI-Native Calibration

Perhaps the most consequential shift is not optical at all. A calibration method called External Variable Augmented Iterative Optimization Technology (EVA-IOT) reduced the number of reference measurements needed to build a robust pharmaceutical powder-stream model by as much as 97%, directly attacking the calibration burden that has long kept process NIR out of smaller manufacturing lines.⁶ Convolutional neural networks are doing similar work upstream, using raw NIR spectra of cell culture media as a checkpoint to flag bioprocess variability before it affects glycosylation outcomes, without requiring a conventional multivariate model to be rebuilt for every new medium lot.⁷ And in antibiotic fermentation, fusing NIR with Raman spectral inputs through a machine-learning layer has enabled genuinely closed-loop control, with the algorithm adjusting feed rates in real time rather than simply reporting a concentration after the fact.⁸ In each case, the instrument's practical resolution and sensitivity increasingly depend as much on the model behind it as on the optics in front of it.

Seeing Around Corners: Standoff and Imaging Applications

Miniaturization and AI are converging most visibly in imaging. NIR hyperspectral imaging paired with convolutional neural network classification has been demonstrated for standoff detection of explosives and other hazardous chemicals in cluttered outdoor scenes, a task that previously required either contact sampling or bulky scanning spectrometers.⁹ A related development applies self-supervised deep learning to markedly shrink the sample sizes needed to train reliable NIR models, addressing one of the field's oldest complaints: that chemometric calibration requires far more reference samples than most laboratories can practically collect.¹⁰ Both point toward NIR sensors that generalize from sparse data rather than demanding an exhaustive calibration set for every new application.

NIR Enters the Clinic, the Cleanroom, and the Fermenter

Portable and embedded NIR platforms are also pushing into biomedical and biomanufacturing settings once reserved for slower reference methods. A comprehensive 2025 review connects the dots between miniaturized, low-cost NIR hardware and applications in pharmaceutical counterfeit detection and point-of-care diagnostics, arguing that accessible instrumentation, not just better chemometrics, is what will determine whether NIR reaches non-expert users at scale.^11,12 Meanwhile, in situ NIR probes monitoring live fermentation are now paired with model-maintenance frameworks that detect and correct calibration drift automatically over weeks of continuous production, addressing a long-standing weakness of process analytical technology: models that work beautifully on day one and quietly degrade by day thirty.¹³

Summary and Conclusions

Across chip-scale photonics, MEMS interferometry, and AI-native calibration, the common thread in recent NIR instrumentation is a redefinition of what counts as “enough” spectral information. Millimeter-scale detector arrays that sample a handful of broad bands are proving sufficient for real quantitative and classification tasks, provided the modeling behind them is strong enough to compensate for what the optics no longer measure.^1,2,6–8 At the same time, ultra-high-resolution photonic chips show that shrinking the package does not have to mean sacrificing performance, at least for well-defined analytes.³ The honest caveat, echoed by researchers who have spent decades separating genuine advances from marketing claims, is that hardware miniaturization has outpaced the guidance, standardization, and user training needed to deploy these instruments safely outside expert hands.⁵

Future Outlook

The near-term trajectory points toward NIR instruments that are simultaneously smaller and more self-sufficient. Expect further convergence of photonic integrated circuits with on-chip machine learning, continued growth of multimodal designs that fuse NIR with Raman or imaging channels for closed-loop process control, and calibration frameworks that maintain themselves over an instrument's operating life rather than requiring periodic manual revalidation.^3,6,8,13 Broader adoption in regulated industries such as biopharmaceutical manufacturing will likely hinge less on further optical innovation than on explainability, uncertainty quantification, and standardized performance benchmarks that let regulators and non-specialist users trust a device whose “spectrum” may now be a handful of AI-interpreted numbers rather than a continuous curve.¹⁴ If that trust gap closes, the next generation of NIR sensing may be defined less by what it measures than by how invisibly it does so.

References

(1) Ou, F.; van Klinken, A.; Hakkel, K. D.; Petruzzella, M.; van Elst, D. M. J.; Ševo, P.; Li, C.; Pagliano, F.; van Veldhoven, R. P. J.; Fiore, A. Spectral Sensing Using a Handheld NIR Module Based on a Fully Integrated Sensor Chip. Spectroscopy 2022, 37 (S11), 34–38. https://doi.org/10.56530/spectroscopy.yd5989g6

(2) Hakkel, K. D.; Petruzzella, M.; Ou, F.; van Klinken, A.; Pagliano, F.; Liu, T.; van Veldhoven, R. P. J.; Fiore, A. Integrated Near-Infrared Spectral Sensing. Nat. Commun. 2022, 13, 103. https://doi.org/10.1038/s41467-021-27662-1

(3) Yao, C.; Zhang, W.; Bao, P.; et al. Chip-Scale Sensor for Spectroscopic Metrology. Nat. Commun. 2024, 15, 10305. https://doi.org/10.1038/s41467-024-54708-x

(4) Miseo, E. V. 2025 Review of Spectroscopic Instrumentation. Spectroscopy 2025, 40 (4), 26–31. https://doi.org/10.56530/spectroscopy.nx9188m9

(5) Yan, H.; Han, B.; Siesler, H. W. Handheld Near-Infrared Spectrometers: Reality and Empty Promises. Spectroscopy 2020, 35 (6), 12–17. Available at https://www.spectroscopyonline.com/view/handheld-near-infrared-spectrometers-reality-and-empty-promises (accessed July 31, 2026).

(6) Velez-Silva, N. L.; Rish, A. J.; Drennen, J. K., III; Anderson, C. A. Robust Near-Infrared Modeling for Pharmaceutical Powder Streams: External Variable Augmented Iterative Optimization Technology (EVA-IOT). Eur. J. Pharm. Biopharm. 2025, 207, 114626. https://doi.org/10.1016/j.ejpb.2025.114626

(7) Gangwar, N.; Balraj, K.; Rathore, A. S. Near-Infrared Spectroscopy Coupled with Convolutional Neural Network as a Checkpoint Tool for Cell Culture Bioprocess Media Characterization. Biotechnol. Prog. 2025, 41 (6), e70056. https://doi.org/10.1002/btpr.70056

(8) Xu, F.; Su, L.; Gao, H.; Wang, Y.; Ben, R.; Hu, K.; Mohsin, A.; Li, C.; Chu, J.; Tian, X. Harnessing Near-Infrared and Raman Spectral Sensing and Artificial Intelligence for Real-Time Monitoring and Precision Control of Bioprocesses. Bioresour. Technol. 2025, 421, 132204. https://doi.org/10.1016/j.biortech.2025.132204

(9) Workman, J., Jr. AI-Powered Near-Infrared Imaging Remotely Identifies Explosives. Spectroscopy Online, June 11, 2025. Available at https://www.spectroscopyonline.com/view/ai-powered-near-infrared-imaging-remotely-identifies-explosives (accessed July 31, 2026).

(10) Workman, J., Jr. The Marriage of Near-Infrared Spectroscopy with AI: The Small-Sample Breakthrough. Spectroscopy Online, January 27, 2025. Available at https://www.spectroscopyonline.com/view/the-marriage-of-near-infrared-spectroscopy-with-ai-the-small-sample-breakthrough (accessed July 31, 2026).

(11) Workman, J., Jr. Advancing Near-Infrared Spectroscopy and Machine Learning for Personalized Medicine. Spectroscopy Online, February 12, 2025. Available at https://www.spectroscopyonline.com/view/advancing-near-infrared-spectroscopy-and-machine-learning-for-personalized-medicine (accessed July 31, 2026).

(12) Kos, J.; Pavelek, D.; Kaykhaii, M.; Olsen, M.; Jampilek, J.; Halko, R. Unveiling the Transformative Power of Near-Infrared Spectroscopy in Biomedical and Pharmaceutical Analysis: Trends, Advancements, and Applications. Eur. J. Pharm. Sci. 2025, 196, 107175. https://doi.org/10.1016/j.ejps.2025.107175

(13) Chandrasekaran, K.; Chandrasekaran, K.; Jayaraman, G.; Bhatt, N. Model Maintenance, Monitoring, and Control Framework Using In Situ Near-Infrared Spectroscopy and Offline Process Data for Lactococcus lactis Fermentation. Chem. Eng. J. 2025, 475, 165116. https://doi.org/10.1016/j.cej.2025.165116

(14) Workman, J., Jr. AI Developments That Changed Vibrational Spectroscopy in 2025. Spectroscopy Online, December 27, 2025. https://doi.org/10.56530/spectroscopy.no4686u3